Characteristics Analysis and Domain-Adaptive Recognition Methodology of Partial Discharge for C₄F₇N/CO₂ Eco-Friendly GIS

• Published in: IEEE transactions on dielectrics and electrical insulation Vol. 31; no. 6; pp. 3100 - 3109
• Main Authors: Li, Zhuoxiao, Zang, Yiming, Li, Ze, Huang, Tiancheng, Sun, Weihao, Xu, Yongpeng, Jiang, Xiuchen
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.12.2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Acoustics; Adaptation; Algorithms; Carbon dioxide; Classification; Classification algorithms; C₄F₇N/CO₂ gas mixture; deep learning; Diagnosis; Dielectrics and electrical insulation; Dimensional analysis; Discharge; Discharges (electric); domain adaptation; Electromagnetic scattering; Gas mixtures; Gases; Insulation; partial discharge (PD); Sulfur hexafluoride; Switchgear
• ISSN: 1070-9878; 1558-4135
• DOI: 10.1109/TDEI.2024.3425315

As an environmentally friendly gas, C4F7N/ CO2 gas mixture is expected to replace SF6 gas as the insulating medium of gas-insulated switchgear (GIS). However, current research on the characteristics analysis and recognition of partial discharge (PD) signals in C4F7N/CO2 gas mixture is still insufficient. Therefore, there is an urgent need to study the characteristics of PD signals within C4F7N/CO2 gas mixture to guide the detection and diagnosis of PD in C4F7N/CO2 gas mixture GIS. This article explores PD signal characteristics and classification methods within C4F7N/CO2 gas mixture. A PD experimental platform is established based on a true-type GIS, and a PD signal recognition dataset is constructed. The correlations and distinctions among PD signals in different gases are elucidated by analyzing the spectral and high-dimensional intermediate features of PD signals. Finally, a domain-adaptation PD recognition model is proposed, requiring only a minimal amount of C4F7N/CO2 gas mixture PD signal data for training. This model solves the problem of the decline in accuracy of the SF6 gas PD classification algorithm on the C4F7N/CO2 gas mixture PD signal due to domain shift, enabling the PD classification algorithm for SF6 gas to also be effective for C4F7N/CO2 gas mixture, significantly enhancing the algorithm's applicability and promoting the use of C4F7N/CO2 gas mixture equipment. The domain-adaptation PD recognition model achieves an accuracy of over 99% for recognizing PD signals in SF6 gas and C4F7N/CO2 gas mixture, providing technical support for PD detection and diagnosis in C4F7N/CO2 gas mixture equipment.